OBJECTIVES:The current study evaluates the efficacy of artificial intelligence (AI)-assisted measurement of cervical length (CL) in predicting spontaneous preterm birth (sPTB), comparing the traditional single-line and two-line methods with the innovative AI-line method in the first trimester of pregnancy. MATERIALS AND METHODS:This study is a retrospective secondary analysis of ultrasound images collected prospectively from women with a viable singleton pregnancy who were undergoing Down syndrome screening at Prince of Wales Hospital, Hong Kong SAR. CL was measured using transvaginal ultrasound, with a secondary analysis of archived 1664 images acquired during a prospective study and processed through a ResUNet-based model. This model, combining UNet and ResNet architectures, a modified ResUNet framework, aimed to overcome the limitations of current measurement techniques by providing a more accurate prediction of CL, particularly in cases where the cervix is curved. RESULTS:The AI-line method demonstrated superior accuracy in predicting sPTB at <37 and <32 weeks of gestation compared with conventional methods, with higher areas under the receiver operating characteristic curve (AUROC). The AUROC of CL measured by the AI-line method (0.676 [95% CI, 0.616-0.735], P < 0.05) in predicting sPTB at <37 weeks of gestation was significantly higher than the single-line (0.537 [95% CI, 0.474-0.6]) and two-line (0.54 [95% CI, 0.473-0.66]) methods. For the prediction of sPTB at <32 weeks of gestation, the AI-line method achieved an AUROC of 0.777 (95% CI, 0.703-0.850). CONCLUSION:The AI-line method offers a more accurate measurement of CL in the first trimester, showing potential as a tool for early screening of sPTB risk. The study's results could significantly influence clinical decision-making, providing a basis for the potential future clinical application of AI in prenatal care.

The CO2 detection is crucial for both industrial and domestic use, as prolonged exposure to high levels can pose serious health risks. Recently, chemiresistive sensors have been demonstrated to be an affordable solution for detecting CO2 in ambient conditions. In this study, a high-sensitivity chemiresistive sensor, consisting of an urchin-like TiO2 microsphere coupled with a 2D MXene layer, was developed for CO2 detection at room temperature over a wide range of 50-5000 ppm. By optimizing different aspects, such as unique urchin-like TiO2/MXene heterostructure, operating temperatures, and relative humidity (%RH), the urchin-like TiO2/MXene heterostructure sensors with a MXene concentration of 5 wt % exhibited high sensitivity and selectivity toward CO2 gas under 50% RH at 30 degrees C temperature. The prepared urchin-like TiO2/MXene-based chemiresistive sensor exhibited a sensing response (R) of 30.69% for 500 ppm of CO2 at 30 degrees C and 50% RH, significantly higher than that of pure TiO2 (SR. 10%) and pure MXene (SR. 5%). Moreover, the gas sensor based on the TiO2/MXene heterostructure demonstrated significantly faster response and recovery times, with values of 82 and 92 s, respectively, compared to the pure TiO2 and pristine MXene. The incorporation of MXene significantly improved sensing performance in terms of high sensitivity, long-term stability, and selectivity. These findings demonstrate the potential of MXene-related gas sensors for detecting food spoilage and other critical applications.

Woven carbon fiber composites are increasingly adopted in advanced structural applications due to their exceptional strength-to-weight ratio and tunable design features. However, high-fidelity simulations of their complex woven architecture are computationally intensive. This study presents a hybrid deep learning framework that combines a dual-input Convolutional Neural Network (CNN) for mechanical property prediction with a Deep Q-Network (DQN) for reinforcement learning-based optimization. The CNN achieves R2 values above 0.96 for elastic deformation, plastic deformation, and strain energy density prediction. Using the DQN, the optimized design achieves a 2.37-fold improvement in strain energy density, increasing from 3590.78 J/m3 to 8527.85 J/ m3. Furthermore, by replacing the original woven geometry with a reduced model using stress-strain behavior, simulation time is reduced from 534 min to 2 min, a 267-fold speedup. This approach significantly enhances efficiency in composite design and optimization workflows, enabling rapid exploration of high-performance configurations.

Flip-Chip Bonding (FCB) is a key packaging technology that shortens interconnects and improves electrical and thermal performance, but mismatches in coefficients of thermal expansion (CTE) often cause substrate warpage during reflow. This deformation leads to solder joint cracking, delamination, and reduced reliability, posing critical challenges in advanced packaging.This study proposes an intelligent optimization framework that integrates finite element modeling (FEM) with reinforcement learning (RL) to design magnetic cover jigs for warpage control. FEM simulations in Abaqus capture thermomechanical responses under realistic thermal loading, while automated scripting provides rapid case generation and analysis. The RL agent explores high-dimensional design spaces and autonomously identifies optimal jig geometries.In addition to minimizing global warpage, a scoring system is introduced to achieve region-specific shaping, enabling complementary warpage profiles that better align with die deformation. Results show that the proposed FEM–RL framework significantly reduces substrate deformation, supports targeted warpage control, and improves packaging reliability while lowering design cost and cycle time.

Bioinspired materials are often modeled after natural structures that have been optimized through evolutionary processes, resulting in exceptional properties such as increased strength, toughness, and resilience. However, the intricate geometries of these materials introduce significant computational challenges, particularly in the context of finite element analysis and structural design. To address these complexities, a novel design framework that integrates reinforcement learning with finite element methods is presented. Specifically, the Deep Q-Network (DQN) algorithm is employed to optimize the design parameters of bionic composite materials. This approach allows for the precise tuning of material properties to meet specific design criteria while enabling real-time performance evaluation during the design process. Compared to conventional methods, the framework offers a substantial reduction in design costs. Moreover, the results show that by strategically combining soft and stiff materials, composite structures can engineered with mechanical performance superior to those composed of a single material. This innovative method provides a highly efficient solution for designing bioinspired composite materials, effectively overcoming the computational challenges associated with their complex geometries.

Carbon monoxide (CO) detection at room temperature is essential for public health and safety due to increasing emissions from industrial and automobile sources. Conventional CuO sensors often suffer from low sensitivity, high detection limits, and slow response, motivating the development of improved sensing materials. In this study, Ti3C2 MXene-decorated porous CuO heterostructures were fabricated via a solvothermal process followed by thermal annealing of Cu-MOF templates. The integration of Ti3C2 MXene with CuO introduces a novel heterostructure design that facilitates charge transfer, thereby enhancing sensing performance. The fabricated CuO/ Ti3C2 nanohybrids exhibited excellent CO sensing performance, with a high response of 9.8 at 10 ppm, an exceptionally low detection limit of 1 ppb, and rapid response/recovery times of 12 s/9 s. In addition, the sensors demonstrated excellent repeatability and superior selectivity compared to pristine CuO sensors. To further improve sensing capability, deep learning models were applied. An LSTM-based classification model achieved outstanding accuracy of 0.989 and 1.0 on the training and test sets, respectively, for concentration prediction. A regression model accurately identified response and recovery times, with average IoU values of 0.84 and 0.81. Cross-validation confirmed the robustness of these models. This combined approach, integrating materials engineering with AI-based predictive modeling, provides a cost-effective and innovative pathway for next-generation, room-temperature CO sensors.

The copolymer, a widely used material in our daily lives, presents a significant challenge in targeted sequence design. While recent advancements in computational simulation and data science offer a promising avenue for addressing this complex issue, challenges persist in labeled data scarcity. In this study, we introduce an uncertainty-based active learning framework for predicting the properties of random copolymers. We found that the active learning strategy allowed for labeling only 40 data points within the design space of 1550 data points, drastically reducing the labeling efforts by 97%. Most data selected by active learning were positioned on the design space’s periphery, transforming the learning task into an interpolation problem. Through integrating active learning and molecular dynamics, we successfully overcame the combinatorial explosion problem in copolymer sequence design, streamlining the data labeling process and culminating in a highly accurate model. This research demonstrates data science’s potential in polymer design, especially when facing data scarcity.

Noble metals are widely recognized for their ability to catalyze the electro-oxidation of organic compounds, with smaller particle sizes significantly enhancing electrocatalytic activity. In this study, catalytic electrodes decorated with atomic-level platinum and Pt-Au clusters were fabricated using cyclic atomic-metal electrodeposition. The interactions between the iminium (protonated imine) groups in emeraldine salt polyaniline (PANI) and metal chloride complexes in the electrolyte enabled precise control over the cluster size and composition. The electrocatalytic activity of these electrodes for propanol oxidation was systematically evaluated using cyclic voltammetry (CV). Notably, PANI electrodes decorated with odd-numbered atomic-level Pt clusters exhibited higher peak oxidation currents compared to even-numbered clusters, revealing a unique even-odd effect. For atomic-level Pt-Au clusters, the catalytic activity was significantly influenced by the sequence of Pt and Au deposition, with PANI-Au1Pt3 achieving the highest catalytic activity (35.34 mA/cm2). Bi-metallic clusters consistently outperformed mono-metallic clusters, and clusters containing only one Pt atom demonstrated superior catalytic activity. These findings provide valuable insights into the design of high-performance catalytic electrodes by leveraging atomic-level control of the cluster size, composition, and deposition sequence, paving the way for advanced applications in electrochemical sensors.

The distribution of precipitate size (PSD) in alloys notably affects the mechanical properties of materials. To explore such PSD-dependent mechanical responses, this study employs the crystal plasticity finite element method (CPFEM), which correlates material hardening with dislocation density. Two distribution types are examined: the normal distribution with varying standard deviations (0 %-30 %) and the Lifshitz-Slyozov-Wagner (LSW) distribution. The analysis reveals that for a normal distribution with constant volume fraction and mean precipitate size, increasing the standard deviation (SD) results in a decrease in yield stress. When the mean size deviates substantially from the critical size, the ultimate tensile stress (UTS) declines as SD increases. Conversely, when the mean size is slightly below the critical size, UTS initially decreases with increasing SD up to 10 % before increasing beyond this threshold. For a mean size equal to the critical size, UTS increases with SD up to 10 % and then decreases. Notably, for mean sizes considerably smaller or larger than the critical size, the stress-strain responses of the LSW distribution and the normal distribution with SD = 20 % are almost identical.

In sawing processes, the manufacturing industry depends on empirical evaluations and operator expertise to manage tool selection, parameter setup, and tool replacement. These traditional methods are inefficient, costly, and inadequate for meeting the growing demands for diverse materials and high-quality standards in modern manufacturing. This study introduces a framework that combines machine learning and physical principles to optimize sawing processes without needing large machining datasets. A columnar sawing machine with an embedded force sensor monitors cutting force in real-time, while offline analyses assess surface roughness and material properties. Datasets were created using carbon steel, die steel, and stainless steel under different machining conditions. Key features were chosen through random forest regression, and a deep-learning model was developed to predict surface roughness based on these physics-guided parameters. Results show that the embedded force sensor effectively captures critical process insights, enables accurate predictions, and helps build a knowledge base for optimizing tool selection, cutting speeds, and surface quality in real-time. This framework works exceptionally well for materials with similar machinability, offering practical guidance for operators. Additionally, integrating real-time monitoring with cloud-based knowledge sharing improves adaptability and scalability across manufacturing scenarios. The study highlights the potential of combining physical insights with machine learning in manufacturing. Future work will expand the framework to encompass a broader range of materials and machining conditions, thereby further enhancing the efficiency and sustainability of manufacturing systems.

To unequivocally distinguish genuine quantumness from classicality, a widely adopted approach focuses on the negative values of a quasi-distribution representation as compelling evidence of nonclassicality. Prominent examples include the dynamical process nonclassicality characterized by the canonical Hamiltonian ensemble representation (CHER) and the nonclassicality of quantum states characterized by the Wigner function. However, to construct a multivariate joint quasi-distribution function with negative values from experimental data is typically highly cumbersome. Here we propose a computational approach utilizing a deep generative model, processing three marginals, to construct the bivariate joint quasi-distribution functions. We first apply our model to tackle the challenging problem of the CHERs, which lacks universal solutions, rendering the problem ground-truth (GT) deficient. To overcome the GT deficiency of the CHER problem, we design optimal synthetic datasets to train our model. While trained with synthetic data, the physics-informed optimization enables our model to capture the detrimental effect of the thermal fluctuations on nonclassicality, which cannot be obtained from any analytical solutions. This underscores the reliability of our approach. This approach also allows us to predict the Wigner functions subject to thermal noises. Our model predicts the Wigner functions with a prominent accuracy by processing three marginals of probability distributions. Our approach also provides a significant reduction of the experimental efforts of constructing the Wigner functions of quantum states, giving rise to an efficient alternative way to realize the quantum state tomography.

Improving the practical uses of a multifunctional humidity sensor requires developing an easy, economical, and environmentally friendly synthesis process. Unfortunately, most humidity sensors have a complicated fabrication process, which drives up their price and restricts their range of applications. In this present work, quantum dots have prevailed as a potential sensing material owing to their small size and large surface area. Herein, we reported the three different colored (green, yellow, and red) based cadmium selenide (CdSe) quantum dots (QDs) using a solution-processed method. Physical characterization of as synthesized CdSe QDs is confirmed using photoluminescence (PL), transmission electron microscopy (TEM), X-ray diffraction (XRD), UV-visible analysis, diffused light scattering (DLS) and Fourier transform infrared spectroscopy (FTIR). TEM analysis of CdSe QDs revealed the average particle size of 6 nm. These CdSe QDs were further employed as capacitive humidity sensors. Among the investigated samples, Cd-1 (prepared by 225 degrees C) exhibited the highest sensitivity 93.842 pF/ % RH with a rapid response and recovery time of 10 s and 13 s, respectively at 20 Hz. The excellent sensitivity of the Cd-1 is accredited to its least particle size and wider energy band gap as compared to Cd-2 and Cd-3 (prepared by 235 and 245 degrees C) samples. Overall, this work opens an avenue for high performance CdSe QDs based humidity sensors.

The current bottleneck in the development of efficient photocatalysts for hydrogen evolution is the limited availability of high-performance acceptor units. Over the past nine years, dibenzo[b,d]thiophene sulfone (DBS) has been the preferred choice for the acceptor unit. Despite extensive exploration of alternative structures as potential replacements for DBS, a superior substitute remains elusive. In this study, a symmetry-breaking strategy was employed on DBS to develop a novel acceptor unit, BBTT-1SO. The asymmetric structure of BBTT-1SO proved beneficial for increasing multiple moment and polarizability. BBTT-1SO-containing polymers showed higher efficiencies for hydrogen evolution than their DBS-containing counterparts by up to 166 %. PBBTT-1SO exhibited an excellent hydrogen evolution rate (HER) of 222.03 mmol g −1  h −1 and an apparent quantum yield of 27.5 % at 500 nm. Transient spectroscopic studies indicated that the BBTT-1SO-based polymers facilitated electron polaron formation, which explains their superior HERs. PBBTT-1SO also showed 14 % higher HER in natural seawater splitting than that in deionized water splitting. Molecular dynamics simulations highlighted the enhanced water-PBBTT-1SO polymer interactions in salt-containing solutions. This study presents a pioneering example of a substitute acceptor unit for DBS in the construction of high-performance photocatalysts for hydrogen evolution.

2D materials such as graphene, monolayer MoS2 and MXene are highly functional for their unique mechanical, thermal and electrical features and are considered building blocks for future ultrathin, flexible electronics. However, they can easily fracture from flaws or defects and thus it is important to increase their toughness in applications. Here, inspired by natural layered composites and architected 3D printed materials of high toughness, we introduce architected defects to the 2D materials and study their fracture in molecular dynamics simulations. We find that the length of the defects in the shape of parallel bridges is crucial to fracture toughness, as long bridges can significantly increase the toughness of graphene and MoS2 but decrease the toughness of MXene, while short bridges show opposite effects. This strategy can increase the toughness of 2D materials without introducing foreign materials or altering the chemistry of the materials, providing a general method to improve their mechanics.

Deep vein thrombosis (DVT) causes significant healthcare burdens worldwide. This study aims to establish a deep learning model for the diagnosis of DVT from the assessment of vein compressibility. Considering the complexity of ultrasound images, convolutional neural networks with UNet and residual neural network (ResNet) are established for image segmentation, from venous duplex ultrasonographic video images, obtained through standard and portable handheld ultrasound methods. To further evaluate the similarity between the predicted and ground truth images, the structural similarity index (SSIM) is employed. Our deep learning model achieves over 90% accuracy, providing an innovative tool for both images and videos. This study harnesses the power of machine learning to develop an automatic labeling tool that can diagnose DVT by analyzing ultrasonography images. To make the tool more accessible to front-line clinicians, a user-friendly application is created to quickly assess possible clinical severity and enable prompt medical intervention, reducing disease progression.

In response to the urgent need for advanced climate change mitigation tools, this study introduces an innovative CO2 gas sensor based on p-p-type heterostructures designed for effective operation at room temperature. This sensor represents a significant step forward, utilizing the synergistic effects of p-p heterojunctions to enhance the effective interfacial area, thereby improving sensitivity. The incorporation of CuO nanoparticles and rGO sheets also optimizes gas transport channels, enhancing the sensor's performance. Our CuO/rGO heterostructures, with 5 wt % rGO, have shown a notable maximum response of 39.6-500 ppm of CO2 at 25 degrees C, and a low detection limit of 2 ppm, indicating their potential as high-performance, room-temperature CO2 sensors. The prepared sensor demonstrates long-term stability, maintaining 98% of its initial performance over a 30-day period when tested at 1-day intervals. Additionally, the sensor remains stable under conditions of over 40% relative humidity. Furthermore, a first-principles study provides insights into the interaction mechanisms with CO2 molecules, enhancing our understanding of the sensor's operation. This research contributes to the development of CO2 monitoring solutions, offering a practical and cost-effective approach to environmental monitoring in the context of global climate change efforts.

Designing de novo enzymes is complex and challenging, especially to maintain the activity. This research focused on motif design to identify the crucial domain in the enzyme and uncovered the protein structure by molecular docking. Therefore, we developed a Generative Redesign in Artificial Computational Enzymology (GRACE), which is an automated workflow for reformation and creation of the de novo enzymes for the first time. GRACE integrated RFdiffusion for structure generation, ProteinMPNN for sequence interpretation, CLEAN for enzyme classification, and followed by solubility analysis and molecular dynamic simulation. As a result, we selected two gene sequences associated with carbonic anhydrase from among 10,000 protein candidates. Experimental validation confirmed that these two novel enzymes, i.e., dCA12_2 and dCA23_1, exhibited favorable solubility, promising substrate-active site interactions, and achieved activity of 400 WAU/mL. This workflow has the potential to greatly streamline experimental efforts in enzyme engineering and unlock new avenues for rational protein design.

Carbon monoxide (CO) is a byproduct of the incomplete combustion of carbon-based fuels, such as wood, coal, gasoline, or natural gas. As incomplete combustion in a fire accident or in an engine, massively produced CO leads to a serious life threat because CO competes with oxygen (O2) binding to hemoglobin and makes people suffer from hypoxia. Although there is hyperbaric O2 therapy for patients with CO poisoning, the nanoscale mechanism of CO dissociation in the O2-rich environment is not completely understood. In this study, we construct the classical force field parameters compatible with the CHARMM for simulating the coordination interactions between hemoglobin, CO, and O2, and use the force field to reveal the impact of O2 on the binding strength between hemoglobin and CO. Density functional theory and Car-Parrinello molecular dynamics simulations are used to obtain the bond energy and equilibrium geometry, and we used machine learning enabled via a feedforward neural network model to obtain the classical force field parameters. We used steered molecular dynamics simulations with a force field to characterize the mechanical strength of the hemoglobin-CO bond before rupture under different simulated O2-rich environments. The results show that as O2 approaches the Fe2+ of heme at a distance smaller than ∼2.8 Å, the coordination bond between CO and Fe2+ is reduced to 50% bond strength in terms of the peak force observed in the rupture process. This weakening effect is also shown by the free energy landscape measured by our metadynamics simulation. Our work suggests that the O2-rich environment around the hemoglobin-CO bond effectively weakens the bonding, so that designing of O2 delivery vector to the site is helpful for alleviating CO binding, which may shed light on de novo drug design for CO poisoning.

Polyethylene terephthalate (PET) is the most abundant plastic waste in the environment. Currently, a new biocatalyst PETase, was discovered in 2016 from Ideonella sakaiensis bacteria, owned the high ability to digest PET through a mild and sustainable process. However, the high-level production of PETase in the model Escherichia coli remains a challenge and limits its application. Therefore, we employed the native molecular chaperones from Ideonella sakaiensis to improve the quality and quantity of an outstanding PETase variant, FASTPETase (FA) at the first time. We selected GroELS from E. coli (EcG) and I. sakaiensis (IsG) using three genetic designs while the co-expressing FA with IsG chaperone increased soluble FA and elevated its activity by 25%. On the other hand, through the genome mining of I. sakaiensis, we identified a lipase secretion chaperone (IsLsC) at the upstream of native PETase. When co-expressing IsLsC and FA, the degradation efficiency toward PET film was up to 51.7 % within one day at 50 degrees C. More LsC-like chaperones could be explored from the sequence similarity network (SSN) with corresponding function to IsLsC. Finally, molecular docking and dynamic simulation exploited a hydrogen bond formation between FA and IsLsC to stabilizing the overall structure. The discovery of a novel chaperone offers a promising strategy for attractive PETase engaging in PET waste valorization.

Mini-emulsion and nanoprecipitation techniques relied on large amounts of surfactants, and unresolved miscibility issues of heterojunction materials limited their efficiency and applicability in the past. Through our molecular design and developed surfactant-free precipitation method, we successfully fabricated the best miscible bulk-heterojunction-particles (BHJP) ever achieved, using donor (PS) and acceptor (PSOS) polymers. The structural similarity ensures optimal miscibility, as supported by the interaction parameter of the PS/PSOS blend is positioned very close to the binodal curve. Experimental studies and molecular dynamics simulations further revealed that surfactants hinder electron output sites and reduce the concentration of sacrificial agents at the interface, slowing polaron formation. Multiscale experiments verified that these BHJP, approximately 12 nm in diameter, further form cross-linked fractal networks of several hundred nanometers. Transient absorption spectroscopy showed that BHJP facilitates polaron formation and electron transfer. Our BHJP demonstrated a superior hydrogen evolution rate (HER) compared to traditional methods. The most active BHJP achieved an HER of 251.2 mmol h-1 g-1 and an apparent quantum yield of 26.2% at 500 nm. This work not only introduces a practical method for preparing BHJP but also offers a new direction for the development of heterojunction materials.